Foundations of Artificial Intelligence 4. Introduction: Environments - PowerPoint PPT Presentation

Foundations of Artificial Intelligence 4. Introduction: Environments and Problem Solving Methods Malte Helmert University of Basel February 25, 2019

Environments Problem Solving Methods Classification of AI Topics Summary Introduction: Overview Chapter overview: introduction 1. What is Artificial Intelligence? 2. AI Past and Present 3. Rational Agents 4. Environments and Problem Solving Methods

Environments Problem Solving Methods Classification of AI Topics Summary Environments of Rational Agents

Environments Problem Solving Methods Classification of AI Topics Summary AI Problems AI Problems AI problem: performance measure + agent model + environment German: Performance-Mass, Agentenmodell, Umgebung agent model: Which actions are at the agent’s disposal? Which observations can it make? environment: Which aspects of the world are relevant for the agent? How does the world react to the agent’s actions? Which observations does it send to the agent?

Environments Problem Solving Methods Classification of AI Topics Summary Example Problem: Autonomous Taxi Example (Autonomous Taxi) environment: streets, vehicles, pedestrians, weather, . . . performance measure: punctuality, safety, profit, legality, comfort, . . . agent model: actions: steering, accelerating, braking, changing gears, honking, . . . observations: cameras, acceleration sensors, GPS, touchpad, . . .

Environments Problem Solving Methods Classification of AI Topics Summary Example Problem: Web Shopping Bot Example (Web Shopping Bot) environment: web pages, products, sellers, . . . performance measure: cost and quality of bought products, shipping time, . . . agent model: actions: querying the user, following links, filling in forms, . . . observations: HTML pages (text, images, scripts, metadata), user input, . . .

Environments Problem Solving Methods Classification of AI Topics Summary Classification of Environments properties of environment determine character of an AI problem classify according to criteria such as: static vs. dynamic deterministic vs. non-deterministic vs. stochastic fully vs. partially vs. not observable discrete vs. continuous single-agent vs. multi-agent

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static deterministic observability discrete agents static vs. dynamic Does the state of the environment change while the agent is contemplating its next action? German: statisch, dynamisch

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic observability discrete agents static vs. dynamic Does the state of the environment change while the agent is contemplating its next action? German: statisch, dynamisch

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic observability discrete agents deterministic vs. non-deterministic vs. stochastic Is the next state of the environment fully determined by the current state and the agent’s next action? If not: is the next state affected by randomness? German: deterministisch, nichtdeterministisch, stochastisch

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability discrete agents deterministic vs. non-deterministic vs. stochastic Is the next state of the environment fully determined by the current state and the agent’s next action? If not: is the next state affected by randomness? German: deterministisch, nichtdeterministisch, stochastisch

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability discrete agents completely vs. partially vs. not observable Do the agent’s observations fully determine the state of the environment? If not: can the agent at least determine some aspects of the state of the environment? German: vollst¨ andig/teilweise/nicht beobachtbar

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability fully fully partially partially discrete agents completely vs. partially vs. not observable Do the agent’s observations fully determine the state of the environment? If not: can the agent at least determine some aspects of the state of the environment? German: vollst¨ andig/teilweise/nicht beobachtbar

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability fully fully partially partially discrete agents discrete vs. continuous Is the environment’s state given by discrete or by continuous parameters? also applies to: actions of the agent, observations, elapsing time German: diskret, stetig

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability fully fully partially partially discrete yes yes yes no agents discrete vs. continuous Is the environment’s state given by discrete or by continuous parameters? also applies to: actions of the agent, observations, elapsing time German: diskret, stetig

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability fully fully partially partially discrete yes yes yes no agents single-agent vs. multi-agent Must other agents be considered? If yes: do the agents behave cooperatively, selfishly, or are they adversaries? German: ein/mehrere Agenten; Gegenspieler

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability fully fully partially partially discrete yes yes yes no agents 1 2 (adversaries) (1) many single-agent vs. multi-agent Must other agents be considered? If yes: do the agents behave cooperatively, selfishly, or are they adversaries? German: ein/mehrere Agenten; Gegenspieler

Environments Problem Solving Methods Classification of AI Topics Summary Properties of Environments Example (Properties of Environments) Rubik’s Cube backgammon shopping bot taxi static yes (yes) (yes) no deterministic yes stochastic (yes) no observability fully fully partially partially discrete yes yes yes no agents 1 2 (adversaries) (1) many suitable problem solving algorithms Environments of different kinds (according to these criteria) usually require different algorithms. The “real world” combines all unpleasant (in the sense of: difficult to handle) properties.

Environments Problem Solving Methods Classification of AI Topics Summary Problem Solving Methods

Environments Problem Solving Methods Classification of AI Topics Summary Three Approaches to Problem Solving We can solve a concrete AI problem (e.g., backgammon) in several ways: Three Problem Solving Methods 1 problem-specific: implement algorithm “by hand” 2 general: create problem description + use general algorithm (solver) 3 learning: learn (aspects of) algorithm from experience German: problemspezifisch, allgemein, lernend all three approaches have strengths and weaknesses (which?) combinations are possible we will mostly focus on general algorithms, but also consider other approaches

Environments Problem Solving Methods Classification of AI Topics Summary General Problem Solvers General problem solving: problem instance = ⇒ language = ⇒ solver = ⇒ solution 1 models to classify, define and understand problems What is a problem instance? What is a solution? What is a good/optimal solution? 2 languages to represent problem instances 3 algorithms to find solutions German: Probleminstanz, Sprache, Solver/L¨ oser, L¨ osung, Modelle

Environments Problem Solving Methods Classification of AI Topics Summary Languages are Key! The Key to General Problem Solving Compactly describe complex models in declarative languages! Two roles for declarative languages: specification: need a description of the model computation: algorithmically exploit problem structure